Dense 3D Reconstruction and Tracking of Dynamic Surface - Advances in Image and Graphics Technologies - page 213

Dense 3D Reconstruction and Tracking of Dynamic
Surface
Jinlong Shi, Suqin Bai, Qiang Qian, Linbin Pang, and Zhi Wang
School of Compute Science and Engineering, Jiangsu University of Science and Technology,
Zhenjiang, China
Abstract. This essay addresses the problem of dense 3D reconstruction and
tracking of dynamic surface from calibrated stereo image sequences. The
primary contribution of this research topic is that a novel framework of 3D
reconstruction and tracking of dynamic surface is proposed, where a surface is
divided into several blocks and block matching in stereo and temporal images is
used instead of matching the whole surface, when all the block correspondences
are obtained, a special bilinear interpolation is applied to precisely reconstruct
and track the integral surface. Performance is evaluated on challenging ground-
truth data generated by 3D max, and then different surface materials, such as
fish surface, paper and cloth are used to test the actual effect. The research
results demonstrate that this framework is an effective and robust method for
dynamic surface reconstruction and tracking.
Keywords: Dense, 3D Reconstruction, Tracking and Dynamic Surface.
1 Introduction
Dense 3D reconstruction and tracking of dynamic surface provides more dynamic
information than reconstruction of static surface; this makes the former more useful in
many applications such as animation, motion capture and medical analysis.
Animations require the real appearance of real-world objects from multi-view video
for simulating the real scenes [1, 4, 22]. When researchers capture and analyze motion
of objects, it is very important to collect accurate data of the distance and orientation
of motion [19]. And in some medical fields, it seems the most practical solution is to
use vision-based techniques for tracking heart motion [20, 8, 11], or establishing
virtual environment of surgeon [14, 13].
However, dense 3D reconstruction and tracking of dynamic surfaces have not
reach the satisfaying level for the current stereo vision methods, and the primary
problem, namely how to precisely perform matching between stereo and temporal
images, is still tough in 3D reconstruction and tracking.
Researchers have proposed a variety of methods for reconstruction of 3D surfaces.
The two types of commonly used methods are marker-based methods [17, 6, 21] and
marker-less methods [18, 12, 7, 5, 3]. Marker-based method uses reflective markers or
special regular textural markers attached to the surface, and track these markers in
calibrated images. But the accuracy is limited by the number of the markers and their